Research map / Label-efficient and robust learning / Test-time adaptation
Von Mises-Fisher Mixture Model with Dynamic Shrinkage for Realistic Test-Time Transduction
Jiazhen Huang, Zhiming Liu, Changhu Wang, Wei Ju, Ziyue Qiao, Xiao Luo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add alpha dynamic shrinkage (ablation) | Ours · Component ablation, bs 64 and 1000 | accuracy: 69.0 -> 72.4 (Bs=64); 68.9 -> 71.1 (Bs=1,000) | Improved |
| add gamma instance weighting (ablation) | Ours · Component ablation, bs 64 and 1000 | accuracy: 72.3 -> 72.4 (Bs=64); 70.4 -> 71.1 (Bs=1,000) | Mixed |
| add iterative update of mu (ablation) | Ours · Component ablation, bs 64 and 1000 | accuracy: 66.8 -> 72.4 (Bs=64); 66.5 -> 71.1 (Bs=1,000) | Improved |
| add iterative update of kappa (ablation) | Ours · Component ablation, bs 64 and 1000 | accuracy: 70.8 -> 72.4 (Bs=64); 68.9 -> 71.1 (Bs=1,000) | Improved |
| set alpha_k = 1/lambda_k instead of lambda_k (ablation) | Ours · Medium, bs 1000, ImageNet | accuracy: 68.5 -> 81.6 | Improved |
| set alpha_k = 1/lambda_k instead of lambda_k (ablation) | Ours · Medium, bs 1000, average | accuracy: 67.6 -> 72.9 | Improved |
| set alpha_k = 1/lambda_k instead of exp(-lambda_k) (ablation) | Ours · Medium, bs 1000, average | accuracy: 68.8 -> 72.9 | Improved |
| use geometric mean for lambda_k instead of arithmetic mean (ablation) | Ours · Extreme imbalance, avg | accuracy: 73.9 -> 73.1 | Hurt |
| use geometric mean for lambda_k instead of arithmetic mean (ablation) | Ours · Mild imbalance, avg | accuracy: 67.8 -> 68.4 | Improved |
| use geometric mean for lambda_k instead of max only (ablation) | Ours · Extreme imbalance, avg | accuracy: 71.4 -> 73.1 | Improved |
| replace GMM in StatA with vMF mixture (ablation) | StatA · Batch adaptation avg, bs 64 | accuracy: 69.01 -> 68.69 | Hurt |
| replace Gaussian with vMF mixture (ablation) | Ours · Batch adaptation avg, bs 64 | accuracy: 71.49 -> 72.36 | No clear effect |
Closest papers by meaning
- Realistic Test-Time Adaptation of Vision-Language Models (CVPR 2025 · similarity 0.88)
- Efficient Test-time Adaptation through Candidate Verification and Divergence Shifts (NeurIPS 2026 · similarity 0.88)
- The Illusion of Progress? A Critical Look at Test-Time Adaptation for Vision-Language Models (NeurIPS 2025 · similarity 0.87)
- A Closer Look at the Few-Shot Adaptation of Large Vision-Language Models (CVPR 2024 · similarity 0.87)
- Frustratingly Easy Test-Time Adaptation of Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Boosting Vision-Language Models with Transduction (NeurIPS 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
